Research map / Label-efficient and robust learning / Domain adaptation and generalization
Mutual Nearest Neighbor Contrast and Hybrid Prototype Self-Training for Universal Domain Adaptation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add mutual nearest neighbors contrast (ablation) | MATHS · VisDA CDA, PDA, ODA, and OPDA | accuracy: 65.2, 64.7, 59.3, and 64.1 vs 72.9, 74.8, 70.6, and 72.8 | Helped |
| Add hybrid prototype self-training (ablation) | MATHS · VisDA CDA, PDA, ODA, and OPDA | accuracy: 71.5, 73.9, 68.5, and 71.2 vs 72.9, 74.8, 70.6, and 72.8 | Helped |
| Use hybrid prototype prediction instead of the source classifier (ablation) | MATHS · VisDA ODA and OPDA | accuracy: 66.4 and 69.7 vs 70.6 and 72.8 | Helped |
| Replace mutual nearest neighbors contrast with MoCo (ablation) | MATHS first stage · VisDA CDA, PDA, ODA, and OPDA | accuracy: 68.5, 67.4, 63.1, and 67.2; about 6% lower than MATHS | Hurt |
| Use MATHS rather than the strongest reported baseline | Office CDA · Office CDA (Avg) | accuracy: 88.9 -> 86.8 | Hurt |
| Use MATHS rather than the strongest reported baseline | OfficeHome CDA · OfficeHome CDA (Avg) | accuracy: 68.1 -> 69.9 | Helped |
| Use MATHS rather than the strongest reported baseline | VisDA CDA · VisDA CDA | accuracy: 74.6 -> 72.9 | Hurt |
| Use MATHS rather than the strongest reported baseline | Office-Caltech PDA · Office-Caltech PDA (Avg) | accuracy: 92.8 -> 87.9 | Hurt |
| Use MATHS rather than the strongest reported baseline | OfficeHome PDA · OfficeHome PDA (Avg) | accuracy: 75.9 -> 72.0 | Hurt |
| Use MATHS rather than the strongest reported baseline | VisDA PDA · VisDA PDA | accuracy: 59.8 -> 74.8 | Helped |
| Use MATHS rather than the strongest reported baseline | Office ODA · Office ODA (Avg) | accuracy: 94.1 -> 95.0 | Helped |
| Use MATHS rather than the strongest reported baseline | OfficeHome ODA · OfficeHome ODA (Avg) | accuracy: 78.1 -> 81.4 | Helped |
Closest papers by meaning
- Tackling Dimensional Collapse toward Comprehensive Universal Domain Adaptation (ICML 2025 · similarity 0.90)
- Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias (CVPR 2024 · similarity 0.90)
- Gradient-Based Sample Selection for Black-Box Universal Domain Adaptation (AAAI 2025 · similarity 0.90)
- MLNet: Mutual Learning Network with Neighborhood Invariance for Universal Domain Adaptation (AAAI 2024 · similarity 0.90)
- Low Category Uncertainty and High Training Potential Instance Learning for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.89)
- SimProF: A Simple Probabilistic Framework for Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.